Insiders Hub

An Efficient Approach for Content Based Image Retreival (CBIR) with Intel AI Vision Kit.

Shriram KV

Shriram KV

Bengaluru, Karnataka

12 0

Users use image search for image retrieval from multifarious domains like crime investigation, fashion, species identification, medicine, etc. The exact need or pattern of the image search and the images sought have been put to use has not been properly analyzed. Many search engines are being developed that are expected to deliver correctly matched results, faster. The success of a search engine depends on its accuracy and speed. In the past, image search engines were developed based on keyword i.e. search for images based on the keyword associated with that image. There are drawbacks in this kind of search and hence a new methodology called Content-Based Image Retrieval (CBIR) was originated. In this method, the query is given as an image and the problem is to retrieve visually similar images from the given image database. This is carried out by employing image processing techniques. The developed Algorithm has been implemented and tested in Intel AI Vision Kit! ...learn more

Project status: Published/In Market

HPC, Artificial Intelligence

Intel Technologies
Intel CPU

Code Samples [1]

Overview / Usage

The system is built to retrieve the most closely matched image from the data set for the input image. The features of the images are compared and then the closest match arrives. A

Methodology / Approach

Step: 1 Input image is to be fed to the system. The system has been designed to be versatile in such a way that most of the image formats will be accepted. A lot of care and effort has been taken to make it versatile.
Step 2: Texture of the input image is obtained and compared with that of the database images. Texture acts as a facilitator to retrieve a set of images which are closer to the fed input image.
Step 3: Histogram to be generated from the results of entropy filtered images. The texture will provide a set of images which are very closer to the input image fed, thus increasing the scope for the accuracy.
Step 4: Entropy of the input is compared with that of the refined database obtained from the above-mentioned steps.
Step 5: SURF based Region of interest is used to select a particular region from the input image and will be compared with the images filtered out from the above-mentioned sequence.

Six Different Combinations of the techniques have been tried as follows to arrive at the best:

  1. Histogram, Texture, Entropy and ROI
  2. Histogram, Entropy, Texture and ROI
  3. Entropy, Histogram, Texture and ROI
  4. Entropy, Texture, Histogram and ROI
  5. Texture, Entropy, Histogram and ROI
  6. Texture, Histogram, Entropy and ROI

Technologies Used

  1. Python
  2. Intel AI Vision Kit
  3. Image Processing

Repository

http://thescipub.com/abstract/10.3844/jcssp.2014.272.284

Collaborators

There are no people to show.

Comments (0)